lesniewski.
qxp 3/25/98 9:11 AM Page 800
Noncommutative
Geometry
Andrew Lesniewski
Noncommutative Spaces
It was noticed a long time ago that various prop-
erties of sets of points can be restated in terms
of properties of certain commutative rings of
functions over those sets. In particular, this ob-
servation proved to be extremely fruitful in al-
gebraic geometry and has led to tremendous
progress in this subject over the past few
decades. In these developments the concept of
a point in a space is secondary and overshadowed
by the algebraic properties of the (sheaves of)
rings of functions on those spaces.
This idea also underlies noncommutative
geometry, a new direction in mathematics initi-
ated by the French mathematician Alain Connes
and outlined in his recent book [3]. In noncom-
mutative geometry one goes one step further: it
is no longer required that the algebra of func-
tions be commutative! Furthermore, while alge-
braic geometry did not entirely rid itself of the
concept of a point, noncommutative geometry
does not use this concept at all. In fact, a point
in a noncommutative space is often a contra-
diction in terms.
One of the sources of noncommutative geom-
etry is the following classic theorem due to
Gelfand and Naimark.
Theorem 1. Let A be a commutative C∗ -alge-
bra, and let M denote the set of maximal ideals
of A . Then, equipped with a natural topology, M
Andrew Lesniewski was associate professor of physics
at Harvard University. Currently he is vice president of
Acadian Asset Management in Boston, MA. His e-mail
address is alesniew@husc.harvard.edu.
Research supported by NSF Grant DMS-94-24344.
800 NOTICES OF THE AMS
is a locally compact space, and A'C 0 (M) , where
C0 (M) denotes the C∗ -algebra of continuous
functions on M vanishing at infinity.
Recall that a C∗ -algebra is an algebra over C ,
equipped with an involutive operation ∗ and a
norm k·k , which satisfies the condition
kSS ∗ k = kSk2 . In other words, the category of
locally compact spaces is equivalent to the cat-
egory of abelian C∗ -algebras. The points of a
topological space can be characterized in purely
algebraic terms as the maximal ideals of an al-
gebra of functions on the space.
Another important source of inspiration for
noncommutative geometry is quantum physics.
It has been known since the heroic days of quan-
tum mechanics (Heisenberg, Born, Jordan,
Schrödinger, Dirac, von Neumann, … ) that or-
dinary concepts of classical mechanics and sym-
plectic geometry do not apply to the subatomic
world. In order to understand the physical phe-
nomena taking place at the atomic scale, one
needs to replace the concepts of classical geom-
etry by other, noncommutative structures. The
notion of a function on phase space needs to be
replaced by an operator acting on a Hilbert space
of states H or a quantum observable . In Dirac’s
parlance, c-numbers get replaced by q-numbers.
This procedure is called quantization.
The simplest example is that of a flat space
R2 which is the phase space of a particle mov-
ing in one dimension. After quantization, the co-
ordinates q and p of a point in R2 are replaced
by operators q and p which obey the Heisenberg-
Born-Jordan commutation relation
(1) [q, p] = i~I,
VOLUME 44, NUMBER 7
lesniewski.qxp 3/25/98 9:11 AM Page 801

where ~ is a fundamental constant of nature, an antihomomorphism S , which satisfies the

Planck’s constant. Explicitly, one takes the Hilbert three conditions stated above. For the case of
space of states to be H = L2 (R) , and SU(2) , the corresponding quantum group (or
q ψ (x) = xψ (x) , pψ (x) = −i~ψ 0 (x). This quanti- q -deformation) is constructed as follows. We
zation procedure results in a structure which can let a, b denote the complex valued functions on
be thought of as a noncommutative deformation SU(2) assigning to a group element g the cor-
of a classical phase space. Heisenberg’s uncer- responding matrix entry in
tainty principle implies that there is no natural
concept of a point on this quantum deformed
³ ´
a b
phase space: all we have is a nonabelian algebra SU(2) 3 g = .
−b a
of “functions on the noncommutative plane”.
One can also quantize classical phase spaces The algebra generated by these functions is
with more complicated geometry; the simplest abelian. The ( C∗ -algebra of functions on the)
of them is the torus T2 . Here the algebra of ob- quantum group SUq (2) is defined as a defor-
servables is generated by two unitary operators mation of the algebra of functions on SU(2).
u and v , such that We will denote the generators of the deformed
algebra by the same symbols, a, b, a, b , and im-
(2) uv = e2π iλ vu, pose the relations
where λ = 2π ~ . One can think of u and v as
ab = qba, ab = qab,
quantizations of the classical functions e2π ip
and e2π iq , respectively. The resulting noncom-
mutative algebra is called the (algebra of func- bb = bb, ba = qab,
tions on the) quantum torus or the irrational ro-  
tation algebra. This algebra also appears ba = qab, aa − aa = q −1 − q bb,
naturally in other contexts (periodic structures
in magnetic fields, matrix models of string the-
where q is a real parameter such that |q| < 1 .
ory). Quantization of more complicated geome-
The abelian case, that of the algebra of con-
tries leads to noncommutative structures which
tinuous functions on SU(2) , corresponds to
cannot be described as easily.
q = 1. For the construction of the relevant Hopf
Related noncommutative structures arise as
algebra structure we refer the reader to the lit-
q -deformations of Lie groups (see, e.g., [5, 8]).
erature.
These structures are often called quantum
Voiculescu’s free probability theory [7] is an-
groups.1 Here the noncommutative algebras carry
other example of a noncommutative structure
the additional structure of a Hopf algebra. This
motivated by physics applications. Here the con-
extra structure on the algebra of functions en-
cept of probability space is replaced by a non-
codes the fact that the underlying noncommu-
commutative structure leading to noncommut-
tative space is a group-like object. To illustrate
ing random variables. One of the main results
this, let us consider the abelian case first. Let F
of this theory, Voiculescu’s central limit theorem,
be an algebra (whose topological structure we ig-
yields the Wigner semicircle law, which arises in
nore) of complex functions on a group G with
the theory of random matrices. Related fields of
the identity element e . For f ∈ F we set
quantum ergodic theory and quantum informa-
∆f (g1 , g2 ) = f (g1 g2 ) , εf (g) = f (e) , and Sf (g) =
tion theory have recently been the focus of a great
f (g −1 ) . This defines algebra homomorphisms
deal of attention. They play a pivotal role in the
∆ : F → F ⊗ F (“coproduct”) and ε : F → C
emerging field of quantum computation.
(“counit”) and an algebra antihomomorphism
Interesting examples of noncommutative
S : F → F (“antipode”). The usual properties defin-
spaces abound, and they are thoroughly dis-
ing a group can equivalently be formulated in
cussed in Connes’ book. In fact, it turns out that
the language of these homomorphisms without
noncommutative geometry also provides a con-
ever referring to the notion of an element of G .
venient framework for studying “commutative”
Explicitly, we verify easily that (∆ ⊗ Id) ◦ ∆ =
but highly singular structures. These include
(Id ⊗ ∆) ◦ ∆ , (ε ⊗ Id) ◦ ∆ = (Id ⊗ ε) ◦ ∆ = Id , and
fractal sets and products of smooth manifolds
m ◦ (S ⊗ Id) ◦ ∆ = Iε . In the last of these identi-
by finite sets.
ties, m denotes the multiplication on F , and I
is the identity element of F . A Hopf algebra is K-Theory and Fredholm Modules
an algebra (not necessarily commutative) which
An important part of Connes’ program is the no-
is equipped with homomorphisms ∆ and ε and
tion of a vector bundle over a noncommutative
1The term “quantum groups” is somewhat misleading, space. The well-known Swan’s theorem states
as these structures are not quantizations of classical that the algebraic K0 group of the algebra C(M)
groups in the physical sense described above. (defined in terms of stable isomorphism classes

AUGUST 1997 NOTICES OF THE AMS 801

lesniewski.qxp 3/25/98 9:11 AM Page 802
of projective C(M)-modules), where M is com-
pact, coincides with the Grothendieck group
K(M) (defined in terms of locally trivial, finite-
dimensional complex vector bundles over M )
of the underlying manifold. In the framework of
noncommutative geometry one can thus regard
the elements of the group K0 (A) , where A is
now a nonabelian algebra of functions on a non-
commutative space, as (equivalence classes of)
vector bundles over the noncommutative space.
The K-theory of operator algebras was orig-
inated by Brown, Douglas, and Fillmore [1] in the
mid-seventies—and, incidentally, they gave the
first look into the Connes program. Their goal
was to reinterpret the classical K-theory in terms
of operator algebras, thus extending K-theory to
general topological spaces. They classified all C∗ -
algebra extensions
1 → K → A → C(M) → 1
of A , where K is an ideal of compact operators
and where C(M) is a commutative C∗ -algebra.
They showed how to construct a group from
such extensions. For M compact and finite di-
mensional, this group is isomorphic to the Steen-
rod K-homology of M , K1 (M) .
One of the fundamental concepts of K-theory
is a Fredholm module over an algebra A . An odd
Fredholm module is a triple (H , π , F) consist-
ing of a Hilbert space H , a ∗ -representation π
of A by bounded linear operators on H , and a
self-adjoint unitary operator F such that
[F, π (a)] is compact for all a ∈ A . A Fredholm
module is called even if, in addition, the Hilbert
space H is Z2 -graded, meaning that H comes
equipped with a self-adjoint unitary operator γ
such that γF + Fγ = 0 .
Fredholm modules arise naturally in ordinary
differential geometry. Let M be a Riemannian
p
manifold, and let Λ2 (M) denote the space of
square L integrable p -forms on M . Set
p
H = p Λ2 (M) , and let γ = (−1)p. With F de-
noting the phase of the de Rham operator
p p+1
d : Λ2 (M) → Λ2 (M) , d = F |d| , the triple
(H , π , F) is an even Fredholm module. Simi-
larly, a Dirac operator D on a spin manifold M
defines a Fredholm module.
Fredholm modules also arise in various con-
texts in physics. Historically, the first example
came up in Dirac’s theory of the electron and,
in fact, had a profound impact on modern index
theory. More recently, speculations about high-
energy physics led to the general concept of su-
persymmetry. Supersymmetry is a deep gener-
alization of the usual space-time symmetries
which treats bosons (particles with integer spins,
for example, photons) and fermions (particles
with half-integer spins, for example, electrons)
on equal footing. Mathematically this amounts
to replacing the concept of a Lie group by a Z2 -
802 NOTICES OF THE
graded concept, that of a Lie supergroup. The lat-
ter is a noncommutative space whose coordinate
algebra contains anticommuting elements. A
particular combination of the generators of su-
persymmetry acting on the Hilbert space of
states of the system is an infinite-dimensional,
Dirac-type operator.
Other examples of Fredholm modules include
models associated with physical phenomena of
quantum Hall effect and quantum chaos.
Indeed, much of classical differential geom-
etry can be encoded into the concept of a Fred-
holm module. The length of a curve on a mani-
fold M is defined in terms of a Riemannian
structure on the manifold. If γ : [0, 1] → M is a
smooth R 1 q curve, then its length is
L(γ) = 0 gµν γ̇ µ γ̇ ν dt , where g is a metric ten-
sor. The distance d(p, q) between two points p
and q on M is given by inf γ L(γ) , where the in-
fimum is taken over all smooth curves γ con-
necting p and q . For M a compact spin mani-
fold with the Dirac operator D , Connes notices
that this distance can be expressed in terms of
a Fredholm module over C(M) : d(p, q) =
supf |f (p) − f (q)|, where the supremum is taken
over all smooth f such that the commutator
[f , D] is bounded and satisfies k[f , D]k∞ ≤ 1 .
Extrapolating this observation to the noncom-
mutative case, we can thus regard a Fredholm
module over an algebra as defining the metric
structure on the underlying noncommutative
space. This fact underlies, among other things,
Connes’s formulation of the standard model of
elementary interactions.
Cyclic Cohomology and Index Theory
One of the cornerstones of noncommutative
geometry is cyclic cohomology. This cohomology
theory is a far-reaching generalization of the
classical de Rham theory. Originally it was de-
veloped by Connes [2] and, independently, by
Boris Tsygan [6] in the early eighties. Cyclic co-
homology is a refinement of Hochschild coho-
mology and is constructed as follows. We begin
by defining the Hochschild cohomology of the al-
gebra A (technically, what I describe below is the
Hochschild cohomology of A with coefficients
in the dual A∗ ).
Let C n (A, A∗ ) be the space of complex-valued,
(n + 1)-linear functionals on A , whose elements
we will denote by f . We consider the operator
b : C n (A, A∗ ) → C n+1 (A, A∗ ) defined by
(bf )(a0 , a1 , . . . , an+1 )
X
n
= (−1)j f (a0 , . . . , aj aj+1 , . . . an+1 )
j=0
+ (−1)n+1 f (an+1 a0 , a1 , . . . , an ).
AMS VOLUME 44, NUMBER 7
lesniewski.qxp 3/25/98 9:11 AM Page 803
One verifies that b ◦ b = 0 , so that b is a
coboundary operator. The cohomology of the
complex (C ∗ (A, A∗ ), b) is called the Hochschild
cohomology of A and is denoted by H ∗ (A, A∗ ) .
In the case of A = C ∞ (M), where M is a smooth
manifold, the Hochschild cohomology groups re-
produce the spaces of de Rham currents on M .
That is in a way a drawback, since one would like
a cohomology scheme which reproduces, in the
commutative case, the de Rham homology.
To construct such a theory, we let Cλn (A) de-
note the space of those cochains f ∈ C n (A, A∗ )
for which the following cyclicity condition holds:
f (an , a0 , . . . , an−1 ) = (−1)n f (a0 , a1 , . . . , an ).
Then (Cλ∗ (A), b) turns out to be a subcomplex of
(C ∗ (A, A∗ ), b) , and its cohomology is called the
cyclic cohomology of A . The cyclic cohomology
groups are denoted by HC n (A) . The theorem
below shows that cyclic cohomology has the de-
sired property.
Theorem 2. Let M be a smooth compact mani-
fold, and let A = C ∞ (M) denote the algebra of
smooth functions on M . Then there is a canon-
ical isomorphism
M Classical field theories do not allow one to de-
HC n (A) ' ker(b) ⊕ Hn−2j (M),
j is not fixed (particles get created or annihilated
where Hk (M) denotes the de Rham homology
of M .
The cyclic cohomologies of a variety of other
noncommutative spaces (quantum tori, quan-
tum groups, …) have been computed.
Suppose now that (H , π , F) is an even Fred-
holm module over A , with Z2 -grading on H de-
fined by γ. Associated with this Fredholm mod-
ule, is a fundamental cyclic cohomology class
ch(F) , called the Chern character. It is given by
ch(F)(a0 , . . . , an )
(3) =tr(γa0 [F, a1 ] . . . [F, an ]).
This general concept of the Chern character
leads to an index theorem which is a profound
extension of the classical Atiyah-Singer index
theorem. The topological part of the index for-
mula is given in terms of a pairing between
ch(F) and a K0 (A) class.
Quantum Field Theory
Before describing the content of the last chap-
ter of Connes’ book, devoted largely to his for-
mulation of the standard model of fundamen-
tal interactions, I would like to present some
issues currently faced by relativistic quantum
field theory.
Historically, quantum field theories arose as
results of “second quantization” of classical sys-
AUGUST 1997
tems. The term “first quantization” (or simply,
quantization) refers to a procedure which leads
from a classical mechanical description of a sys-
tem to a quantum mechanical description. The
latter is usually formulated in terms of a fun-
damental partial differential equation (“the wave
equation”) involving a (scalar, vector, spinor, …)
field over the space-time and thus may be
thought of as classical field theory. Planck’s
constant ~ enters the wave equation, and the
classical theory is recovered as a suitable limit
as ~ → 0. The most famous example of this pro-
cedure is Schrödinger’s quantization of New-
tonian mechanics, with the wave equation bear-
ing his name. The first quantization of
electromagnetism was carried through by
Maxwell in the nineteenth century. His theory
marked a transition from corpuscular to wave
description of light (the absence of Planck’s con-
stant in that theory is a fluke due entirely to the
masslessness of the photon). Unlike
Schrödinger’s theory, Maxwell’s theory is rela-
tivistic, meaning that it is compatible with the
special theory of relativity. Another famous ex-
ample of a classical field theory is Dirac’s quan-
tization of the electron.
scribe systems in which the number of particles
as a result of interaction), and thus they fail to
apply to high energy physics. A procedure lead-
ing to a description of systems with a variable
number of particles is called “second quantiza-
tion” because of its similarity to quantization.
The starting point of second quantization is a
first quantized (classical) field theory, and the
result is a quantized field theory. If the classi-
cal field theory is relativistic, the resulting quan-
tum field theory is called relativistic quantum
field theory. The table below summarizes these
remarks.
(4)
NOTICES OF THE AMS 803
lesniewski.qxp 3/25/98 9:11 AM Page 804
Shortly after relativistic quantum field theory
was discovered, it became clear that it suffered
from serious conceptual and technical defi-
ciencies. The only known way of extracting any
information from the fundamental equations
was to expand in a power series in the coupling
constant. Such an expansion is called perturba-
tion theory. Unfortunately, it turns out that each
term of perturbation theory, except for the lead-
ing order contribution, is singular, and so the
power series as it stands is meaningless! The
path-breaking work of Feynman, Schwinger,
Dyson, and others resulted in a procedure of ex-
tracting finite parts of the singular expressions
encountered in perturbation theory known as
renormalization theory. Renormalization theory
removes singularities from perturbation theory
at the expense of introducing a number of arbi-
trary constants whose values should be deter-
mined by experiment. That leads one to the re-
quirement that the number of such constants
should be finite (otherwise one could “explain”
any experiment) and that they should be mea-
surable parameters of the theory. Any quantum
field theory satisfying these requirements is
called renormalizable. These concepts led to
some of the most remarkable developments in
physics. The theory of interacting electrons and
photons, quantum electrodynamics, turns out to
be renormalizable and leads to fantastic agree-
ment with experiment.
The requirement of renormalizability became
a paradigm of quantum field theory, and it
proved extremely fruitful. Guided by it, particle
physicists generalized quantum electrodynam-
ics to include other types of interactions: the
Weinberg-Salam model unifying electromagnetic
and weak interactions and the standard model
unifying electromagnetic, weak, and strong in-
teractions. As of today there is a consensus that
the standard model is the correct theory of ele-
mentary interactions. One disturbing fact about
this theory is that gravity has so far resisted in-
clusion into the framework of renormalizable
quantum field theory.
This renormalizability paradigm underwent
a dramatic revision in the seventies as a result
of Wilson’s renormalization group theory. Ac-
cording to Wilson, we do not need to know the
details of the “true” theory of elementary in-
teractions which is valid at all energy scales. It
may as well be that the fundamental theory is
not a quantum field theory at all. All we have
at our disposal is an effective theory, a low en-
ergy limit of this fundamental theory. The con-
cept of renormalizability thus acquires a new
meaning: renormalizable theories, rather than
being “fundamental”, are merely those theories
which survive the scaling down from the fun-
damental scale to the “laboratory” scale. Non-
804 NOTICES OF THE AMS
renormalizable theories get wiped out in the
process of taking this limit.
The Standard Model à la Connes
The last chapter of Connes’ book contains an ac-
count of his formulation of the standard model
of fundamental interactions. Anybody who has
studied the standard model in its usual formu-
lation with all of its plethora of fields and mech-
anisms will appreciate the compact geometric
formulation presented in the book. Connes’ for-
mulation provides a natural geometric principle
underlying this theory within the framework of
noncommutative geometry.
The first basic idea of Connes’ theory, in its
simplest version (which does not yield the full
standard model), consists of the following. Con-
sider a two-sheeted space time R4 × Z2 , where
the R4 factor is the usual space time and the Z2
factor is a “discrete dimension” (this resembles
slightly the Kaluza-Klein theory unifying Ein-
steinian gravity and electromagnetism). The al-
gebra of smooth functions on this space,
C ∞ (R4 ) ⊕ C ∞ (R4 ) , is abelian. However, the de-
rivative in the discrete direction is a finite dif-
ference quotient, and so a Dirac operator D on
this space is somewhat unusual. More refined
versions of the theory (to reproduce the correct
quark content of the standard model) involve
multiplying R4 by more complicated finite sets
and setting up a sophisticated bimodule struc-
ture over the algebra of functions on the re-
sulting product space.
This is where the formalism of noncommu-
tative geometry becomes crucial. Using his al-
gebraic framework, Connes develops the relevant
gauge theory associated with (the Fredholm
module defined by) D . This leads to a natural
concept of the connection form A , curvature
form F , etc.
The second basic idea in Connes’ approach is
the use of the Dixmier trace as the fundamen-
tal functional to define the action of the theory.
The Dixmier trace of a positive operator S with
discrete spectrum λj is given by
1 X
trω (S) = lim λj
N→∞ log N
1≤j≤N
(so it is nontrivial only for operators with loga-
rithmically divergent traces). The Yang-Mills ac-
tion functional of the theory is now given by
trω (|D|−4 F 2 ) . The remarkable fact is that this
action functional plus the fermionic action func-
tional which I am not discussing here reproduce
the action functional of the standard model. To
acknowledge the depth of this fact, one needs
to go through a truly excruciating series of al-
gebraic manipulations.
VOLUME 44, NUMBER 7
lesniewski.qxp 3/25/98 9:11 AM Page 805

Connes’ theory is purely classical, and so it [5] V. G. Drinfeld, Quantum groups, Proceedings of
belongs to the middle column of the table (4). the International Congress of Mathematicians,
He does not propose a quantization scheme 1987, pp. 798–820.
which would be intrinsic to it; the only known [6] B. Tsygan, Homology of matrix algebras over
rings and the Hochschild cohomology, Uspekhi
way to obtain a quantum field theory out of
Mat. Nauk 38 (1983), 217–218.
Connes’ model is to follow the usual rules of [7] D. Voiculescu (ed.), Free probability theory, Amer.
quantum field theory. Because of its classical Math. Soc., Providence, RI, 1996.
character, the model (or its more recent refine- [8] S. L. Woronowicz, Twisted SU(2) group. An ex-
ments, which include Einsteinian gravity [4]) ample of a noncommutative differential calculus,
does not resolve the issues described in the pre- Publ. Res. Inst. Math. Sci. 23 (1987), 613–665.
vious section, and it does not provide any new
physical principle toward this goal. No unified
theory of fundamental interactions is known;
the consensus among physicists is that string
theory is currently the only promising attempt
at such a theory. Despite this, the Connes for-
mulation of the standard model is a truly re-
markable construction and can be regarded as
a major triumph of noncommutative geometry.
Let me mention that some recent develop-
ments in string theory show some striking and
mysterious similarities between Connes’ theory
and the theory of D-branes. Is it possible that
noncommutative geometry underlies the fun-
damentals of string theory?

Final Remarks
It is an impossible task for me to discuss in a
short review the wealth of fascinating math-
ematical phenomena described by Alain Connes.
I focused on the aspects of noncommutative
geometry which I understand best, and these are
its relationships with theoretical physics.
Connes’ Noncommutative geometry is one of
the milestones of mathematics. It lays the foun-
dations of a new branch of mathematics whose
importance is difficult to overestimate. Its im-
pact will be felt by generations of mathematicians
to come, the way Riemann’s Über die Hypothe-
sen … influenced the development of differen-
tial geometry.
The book has a largely programmatic, ex-
pository character: few things are proved, but the
presentation is extremely lucid. It is a source of
ideas, inspiration, ingenious calculations, and
facts for researchers who are interested in one
of the most fascinating developments in math-
ematics.

References
[1] L.L. Brown, R.G. Douglas, and P.A. Fillmore, Ex-
tensions of C∗ -algebras and K-homology,
Ann. Math. 105 (1977), 265–324.
[2] A. Connes, Noncommutative differential geome-
try, Inst. Hautes Études Sci. Publ. Math. 62 (1986),
257–360.
[3] ———, Noncommutative geometry, Academic Press,
1994.
[4] ———, Gravity coupled with matter and the foun-
dation of noncommutative geometry, Comm. Math.
Phys. 182 (1996), 155–176.

AUGUST 1997 NOTICES OF THE AMS 805